Runlayer: Powering the Next Wave of Secure, Enterprise-Ready Agents

Enterprises are at the edge of a revolution. The arrival of AI agents promises new levels of productivity: code writing itself, tickets closing automatically, databases answering their own queries. 

The Model Context Protocol (MCP) opened the door to a new generation of intelligent workflows and agents, but until now, enterprises couldn’t walk through it. By design, MCP connects agents directly to the systems, data, and tools they need to take meaningful action. The challenge is that those same connections create real complexity around infrastructure and security.

Enterprise teams eager to embrace agents and MCP, however, often meet with increased caution. There’s a sense that open-source agentic infrastructure like MCP isn’t capable of improving performance and abiding by security protocols. These concerns are rooted in a fear that enterprises are rolling out AI faster than they can secure it, creating cracks in the defensive wall just as their most sensitive systems become accessible to autonomous code.

Those fears are grounded in some truth: traditional identity and access frameworks aren’t keeping up with agents. They were built for predictable actors: people following known business processes, or static services executing predetermined logic. These paradigms assume finite, known behaviors. But with LLM-driven agents, that model breaks down. These agents are emergent, runtime-driven, goal-based, and context-sensitive. With an agent, there’s no fixed script and no pre-enumerated permission set. Attempting to assign broad static permissions to such an unpredictable system creates a significant security liability.

Today, we’re announcing our investment in Runlayer, the first platform purpose-built to make MCP enterprise-ready. Runlayer unlocks all of the performance gains of agents and ease of MCP without compromising on security. Enterprise customers can deploy Runlayer behind their own VPC, solving not only integration blockers but also unlocking native connections with identity providers like Okta and Entra. Admins get a curated catalog of pre-vetted MCP servers, granular drag-and-drop policy controls, sophisticated approval workflows, and comprehensive dashboards with unified observability across every MCP in their organization. PII and token masking occur automatically, with configurable filters that ensure sensitive data is never left exposed.There are also MCP-specific security models designed to address recent threats that traditional LLM guardrails don’t effectively protect against. The drag-and-drop policy creator lets anyone in the organization—not just security engineers—implement controls aligned with best practices.

Runlayer’s a win for everyone: engineers are faster with MCP seamlessly wired into their stack, while IT and security teams gain visibility and peace of mind. 

While we’ve been tracking the challenges – and opportunities – presented by agent identity management for months, we were still struck by how exceptionally well-poised the Runlayer team is to make MCP enterprise-ready. They’re working with the co-creator of MCP, David Soria Parra. Runlayer’s founders have shaped some of the fastest-growing AI products with  OpenAI and Anthropic, and shipped Zapier’s MCP, Agents, and AI Actions to millions. CEO Andrew Berman brings deep go-to-market experience from his role as CEO of Vowel (acquired by Zapier), while co-founders Tal Peretz and Vitor Balacco bring deep expertise in AI ML engineering and security. This mix of commercial savvy and technical depth has enabled Runlayer to resonate with sophisticated, security-conscious customers like Gusto, Opendoor, and dbt Labs from day one.

We’re excited for the road ahead and to partner with this exceptional team securely enabling AI.